@sema-agent/cli 1.0.42 → 1.0.43

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "@sema-agent/cli",
-  "version": "1.0.42",
+  "version": "1.0.43",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@sema-agent/cli",
-      "version": "1.0.42",
+      "version": "1.0.43",
       "license": "BUSL-1.1",
       "dependencies": {
         "@sema-agent/sdk": "0.0.63",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sema-agent/cli",
-  "version": "1.0.42",
+  "version": "1.0.43",
   "description": "Sema — your own Claude Code-grade coding agent, in the terminal.",
   "license": "BUSL-1.1",
   "bin": {

package/sema.js

@@ -244614,13 +244614,17 @@ __export(engineSideChannel_exports, {
 });
 import { readFileSync as readFileSync23 } from "node:fs";
 import { join as join73 } from "node:path";
-async function sideQueryCapable(client4, baseUrl, signal) {
+async function sideQueryCapable(client4, baseUrl, budget, signal) {
   const hit = sideQueryCapByBase.get(baseUrl);
-  if (hit !== void 0) return hit;
+  if (hit && (hit.ok || Date.now() - hit.at < CAP_FALSE_RETRY_TTL_MS)) return hit.ok;
   try {
-    const caps = await withTimeoutSignal(CAPS_PROBE_TIMEOUT_MS, signal, (s) => client4.capabilities({ signal: s }));
+    const caps = await withTimeoutSignal(
+      Math.min(CAPS_PROBE_TIMEOUT_MS, budget.remaining()),
+      signal,
+      (s) => client4.capabilities({ signal: s })
+    );
     const ok2 = caps.sideQuery === true;
-    sideQueryCapByBase.set(baseUrl, ok2);
+    sideQueryCapByBase.set(baseUrl, { ok: ok2, at: Date.now() });
     return ok2;
   } catch {
     return false;
@@ -244686,7 +244690,11 @@ async function resolvePoolModelRef(callerModel) {
     return void 0;
   }
 }
-async function engineUtilityText(opts) {
+function makeBudget(totalMs) {
+  const start = Date.now();
+  return { remaining: () => Math.max(1, totalMs - (Date.now() - start)) };
+}
+async function engineUtilityCall(opts) {
   const baseUrl = process.env.SEMA_LIVE_BASEURL;
   if (!baseUrl) return null;
   const token = resolveEngineToken();
@@ -244696,117 +244704,90 @@ async function engineUtilityText(opts) {
     principal: process.env.SEMA_LIVE_PRINCIPAL ?? "anon:shell-live"
   });
   if (!client4) throw new Error("sema engine side-channel: wire client unavailable");
-  const cheap = process.env.MODEL_CHEAP_ID?.trim() || void 0;
-  const timeoutMs = opts.timeoutMs ?? SIDE_CHANNEL_TIMEOUT_MS;
+  const cheap = opts.cheapSlot ? process.env.MODEL_CHEAP_ID?.trim() || void 0 : void 0;
+  const budget = makeBudget(opts.timeoutMs ?? SIDE_CHANNEL_TIMEOUT_MS);
+  const firstModel = opts.modelRef ?? cheap;
   const statusOf2 = (e) => {
     const s = e?.status;
     return typeof s === "number" ? s : void 0;
   };
-  const firstModel = opts.modelRef ?? cheap;
-  if (await sideQueryCapable(client4, baseUrl, opts.signal)) {
-    const query2 = (model) => withTimeoutSignal(
-      timeoutMs,
+  const tasksArm = async () => {
+    const submit = (model) => withTimeoutSignal(
+      budget.remaining(),
       opts.signal,
-      (signal) => client4.sideQuery.query(
+      (signal) => client4.tasks.submit(
         {
-          messages: [{ role: "user", content: opts.objective }],
+          objective: opts.objective,
+          sessionId: mintSideSessionId(),
+          // 独立 session:utility 提示词绝不进主会话历史
           ...model ? { model } : {},
-          ...opts.maxTokens ? { maxOutputTokens: opts.maxTokens } : {}
+          ...opts.maxTokens ? { maxTokens: opts.maxTokens } : {},
+          ...opts.systemPrompt ? { systemPrompt: opts.systemPrompt } : {},
+          ...opts.excludeTools ? { excludeTools: [...opts.excludeTools] } : {}
         },
-        { signal }
+        { signal, idempotencyKey: null }
       )
     );
-    let r;
+    let data;
     try {
-      r = await query2(firstModel);
+      data = await submit(firstModel);
     } catch (e) {
       const status3 = statusOf2(e);
       if (opts.noFallback || status3 === void 0 || !firstModel || !CATALOG_GATE_STATUSES.has(status3)) throw e;
-      logForDebugging(`[side-channel] side-query ${status3} with model=${firstModel} \u2014 retrying on the default model`);
-      r = await query2(void 0);
-    }
-    return sideQueryTextOf(r);
-  }
-  const submit = (model) => withTimeoutSignal(
-    timeoutMs,
-    opts.signal,
-    (signal) => client4.tasks.submit(
-      {
-        objective: opts.objective,
-        sessionId: mintSideSessionId(),
-        // 独立 session:utility 提示词绝不进主会话历史
-        ...model ? { model } : {},
-        ...opts.maxTokens ? { maxTokens: opts.maxTokens } : {}
-      },
-      { signal, idempotencyKey: null }
-    )
-  );
-  let data;
-  try {
-    data = await submit(firstModel);
-  } catch (e) {
-    const status3 = statusOf2(e);
-    if (opts.noFallback || status3 === void 0 || !firstModel || !CATALOG_GATE_STATUSES.has(status3)) throw e;
-    logForDebugging(`[side-channel] engine ${status3} with model=${firstModel} \u2014 retrying on the default model`);
-    data = await submit(void 0);
-  }
-  const text2 = typeof data.result === "string" ? data.result.trim() : "";
-  if (!text2) throw new Error("sema engine side-channel: empty result");
-  return text2;
-}
-async function engineUtilityTask(opts) {
-  const baseUrl = process.env.SEMA_LIVE_BASEURL;
-  if (!baseUrl) throw new Error("sema engine side-channel: not active");
-  const token = resolveEngineToken();
-  const client4 = makeEngineWireClient({
-    baseUrl,
-    ...token ? { token } : {},
-    principal: process.env.SEMA_LIVE_PRINCIPAL ?? "anon:shell-live"
-  });
-  if (!client4) throw new Error("sema engine side-channel: wire client unavailable");
-  const timeoutMs = opts.timeoutMs ?? SIDE_CHANNEL_TIMEOUT_MS;
-  if (await sideQueryCapable(client4, baseUrl, opts.signal)) {
-    const r = await withTimeoutSignal(
-      timeoutMs,
+      logForDebugging(`[side-channel] engine ${status3} with model=${firstModel} \u2014 retrying on the default model`);
+      data = await submit(void 0);
+    }
+    const text2 = typeof data.result === "string" ? data.result.trim() : "";
+    if (!text2) throw new Error("sema engine side-channel: empty result");
+    const st2 = data.stats;
+    const usage = st2 ? {
+      input_tokens: st2.promptTokens ?? st2.totalInputTokens ?? 0,
+      output_tokens: st2.outputTokens ?? 0,
+      cache_creation_input_tokens: st2.cacheWriteTokens ?? 0,
+      cache_read_input_tokens: st2.cachedTokens ?? 0
+    } : null;
+    return { text: text2, usage };
+  };
+  if (await sideQueryCapable(client4, baseUrl, budget, opts.signal)) {
+    const query2 = (model) => withTimeoutSignal(
+      budget.remaining(),
       opts.signal,
       (signal) => client4.sideQuery.query(
         {
           messages: [{ role: "user", content: opts.objective }],
-          ...opts.modelRef ? { model: opts.modelRef } : {},
+          ...model ? { model } : {},
           ...opts.maxTokens ? { maxOutputTokens: opts.maxTokens } : {},
           ...opts.systemPrompt ? { systemPrompt: opts.systemPrompt } : {}
         },
         { signal }
       )
     );
+    let r;
+    try {
+      r = await query2(firstModel);
+    } catch (e) {
+      const status3 = statusOf2(e);
+      if (status3 !== void 0 && ENDPOINT_GONE_STATUSES.has(status3)) {
+        sideQueryCapByBase.set(baseUrl, { ok: false, at: Date.now() });
+        logForDebugging(`[side-channel] side-query endpoint gone (${status3}) \u2014 falling back to tasks`);
+        return await tasksArm();
+      }
+      if (opts.noFallback || status3 === void 0 || !firstModel || !SIDEQUERY_MODEL_GATE_STATUSES.has(status3)) throw e;
+      logForDebugging(`[side-channel] side-query ${status3} with model=${firstModel} \u2014 retrying on the default model`);
+      r = await query2(void 0);
+    }
     return { text: sideQueryTextOf(r), usage: usageFromSideQuery(r) };
   }
-  const data = await withTimeoutSignal(
-    timeoutMs,
-    opts.signal,
-    (signal) => client4.tasks.submit(
-      {
-        objective: opts.objective,
-        sessionId: mintSideSessionId(),
-        // 独立 session:utility 提示词绝不进主会话历史
-        ...opts.modelRef ? { model: opts.modelRef } : {},
-        ...opts.maxTokens ? { maxTokens: opts.maxTokens } : {},
-        ...opts.systemPrompt ? { systemPrompt: opts.systemPrompt } : {},
-        ...opts.excludeTools ? { excludeTools: [...opts.excludeTools] } : {}
-      },
-      { signal, idempotencyKey: null }
-    )
-  );
-  const text2 = typeof data.result === "string" ? data.result.trim() : "";
-  if (!text2) throw new Error("sema engine side-channel: empty result");
-  const st2 = data.stats;
-  const usage = st2 ? {
-    input_tokens: st2.promptTokens ?? st2.totalInputTokens ?? 0,
-    output_tokens: st2.outputTokens ?? 0,
-    cache_creation_input_tokens: st2.cacheWriteTokens ?? 0,
-    cache_read_input_tokens: st2.cachedTokens ?? 0
-  } : null;
-  return { text: text2, usage };
+  return await tasksArm();
+}
+async function engineUtilityText(opts) {
+  const r = await engineUtilityCall({ ...opts, cheapSlot: true });
+  return r === null ? null : r.text;
+}
+async function engineUtilityTask(opts) {
+  const r = await engineUtilityCall(opts);
+  if (r === null) throw new Error("sema engine side-channel: not active");
+  return r;
 }
 function extractJsonObjectText(text2) {
   const m2 = text2.match(/\{[\s\S]*\}/);
@@ -244824,7 +244805,7 @@ function jsonDisciplineSuffix(schema) {
 Return ONLY a JSON object (no prose, no markdown code fences) that matches this JSON Schema:
 ${JSON.stringify(schema)}`;
 }
-var SIDE_CHANNEL_TIMEOUT_MS, sideQueryCapByBase, CAPS_PROBE_TIMEOUT_MS, CATALOG_GATE_STATUSES, UTILITY_EXCLUDE_ALL_TOOLS;
+var SIDE_CHANNEL_TIMEOUT_MS, sideQueryCapByBase, CAPS_PROBE_TIMEOUT_MS, CAP_FALSE_RETRY_TTL_MS, CATALOG_GATE_STATUSES, SIDEQUERY_MODEL_GATE_STATUSES, ENDPOINT_GONE_STATUSES, UTILITY_EXCLUDE_ALL_TOOLS;
 var init_engineSideChannel = __esm({
   "build-src/src/sema/engineSideChannel.ts"() {
     init_debug();
@@ -244833,7 +244814,10 @@ var init_engineSideChannel = __esm({
     SIDE_CHANNEL_TIMEOUT_MS = 6e4;
     sideQueryCapByBase = /* @__PURE__ */ new Map();
     CAPS_PROBE_TIMEOUT_MS = 5e3;
+    CAP_FALSE_RETRY_TTL_MS = 5 * 6e4;
     CATALOG_GATE_STATUSES = /* @__PURE__ */ new Set([400, 404, 422]);
+    SIDEQUERY_MODEL_GATE_STATUSES = /* @__PURE__ */ new Set([400, 422]);
+    ENDPOINT_GONE_STATUSES = /* @__PURE__ */ new Set([404, 405, 501]);
     UTILITY_EXCLUDE_ALL_TOOLS = [
       "Bash",
       "PowerShell",
@@ -272431,14 +272415,16 @@ async function semaSideQueryViaEngine(opts, side) {
   else if (opts.output_format?.schema) objective += side.jsonDisciplineSuffix(opts.output_format.schema);
   const isModelValidation = opts.querySource === "model_validation";
   const modelRef = await side.resolvePoolModelRef(opts.model) ?? (isModelValidation ? opts.model : void 0);
-  const text2 = await side.engineUtilityText({
+  const r = await side.engineUtilityTask({
     objective,
     signal: opts.signal,
     // validateModel 类 1-token ping 在任务语义下会空转 → 托底 16,verdict 仍由成功/失败承载
     maxTokens: Math.max(16, opts.max_tokens ?? 1024),
     modelRef,
-    noFallback: isModelValidation
-  }) ?? "";
+    noFallback: isModelValidation,
+    cheapSlot: true
+  });
+  const text2 = r.text;
   const stamp2 = Date.now().toString(36);
   let content;
   if (forcedTool) {
@@ -272465,7 +272451,8 @@ async function semaSideQueryViaEngine(opts, side) {
     content,
     stop_reason: "end_turn",
     stop_sequence: null,
-    usage: {
+    // 对抗复审批 #3:verb 臂权威 usage / tasks 臂 TaskStats 映射;引擎无账时诚实零
+    usage: r.usage ?? {
       input_tokens: 0,
       output_tokens: 0,
       cache_creation_input_tokens: 0,
@@ -355994,7 +355981,7 @@ function useClaudeAiLimits() {
 var import_react64;
 var init_claudeAiLimitsHook = __esm({
   "build-src/src/services/claudeAiLimitsHook.ts"() {
-    import_react64 = __toESM(require_react(), 1);
+    import_react64 = __toESM(require_react());
     init_claudeAiLimits();
   }
 });
@@ -451647,7 +451634,7 @@ function Settings({
 var import_react113, import_jsx_runtime193;
 var init_cmd_config = __esm({
   "build-src/src/sema/overrides/cmd-config.tsx"() {
-    import_react113 = __toESM(require_react(), 1);
+    import_react113 = __toESM(require_react());
     init_useKeybinding();
     init_useExitOnCtrlCDWithKeybindings();
     init_useTerminalSize();
@@ -451658,7 +451645,7 @@ var init_cmd_config = __esm({
     init_Config();
     init_settings_usage();
     init_Stats();
-    import_jsx_runtime193 = __toESM(require_jsx_runtime(), 1);
+    import_jsx_runtime193 = __toESM(require_jsx_runtime());
   }
 });
 
@@ -480451,13 +480438,13 @@ var init_sema_brand = __esm({
         _doc: "Rendered as the 'What's new' feed (createWhatsNewFeed). Top version MUST be >= current app version (2.1.187) and newer than config.lastReleaseNotesSeen so the notes display and full-logo mode triggers. Pinned == app version (2.1.187) + lastReleaseNotesSeen (loginState.ts) so getRecentReleaseNotes()===[] \u2192 condensed logo (matches refs/chrome.empty-input.json).",
         version: "2.1.187",
         notes: [
-          "/btw and background utilities now use the engine's dedicated side-query channel \u2014 no session storage residue at all (previously each call left a small side-session file)",
-          "Utility calls on the new channel have no tool surface whatsoever \u2014 isolation is structural, not instruction-based",
-          "Older engines are detected automatically and keep the previous isolated-session behavior"
+          "Utility channel now adapts live to engine upgrades/rollbacks \u2014 no restarts needed, no stuck calls",
+          "Background utility calls enforce a single strict time budget (probe + retry + transport)",
+          "Token usage from background utilities is now reported accurately (no more zeros in accounting)"
         ]
       },
-      productVersion: "1.0.42",
-      announcement: "sema 1.0.42 \u2014 background utility calls (/btw, session titles, recaps) now run through the engine's new side-query channel: zero session files left behind, tool access structurally impossible, and accurate token accounting."
+      productVersion: "1.0.43",
+      announcement: "sema 1.0.43 \u2014 hardening for the new side-query channel: engine rollbacks/upgrades are now detected live (no stuck utilities), strict time budgets, and real token accounting flows through every background utility call."
     };
   }
 });
@@ -483386,13 +483373,13 @@ var require_sema_brand = __commonJS({
         _doc: "Rendered as the 'What's new' feed (createWhatsNewFeed). Top version MUST be >= current app version (2.1.187) and newer than config.lastReleaseNotesSeen so the notes display and full-logo mode triggers. Pinned == app version (2.1.187) + lastReleaseNotesSeen (loginState.ts) so getRecentReleaseNotes()===[] \u2192 condensed logo (matches refs/chrome.empty-input.json).",
         version: "2.1.187",
         notes: [
-          "/btw and background utilities now use the engine's dedicated side-query channel \u2014 no session storage residue at all (previously each call left a small side-session file)",
-          "Utility calls on the new channel have no tool surface whatsoever \u2014 isolation is structural, not instruction-based",
-          "Older engines are detected automatically and keep the previous isolated-session behavior"
+          "Utility channel now adapts live to engine upgrades/rollbacks \u2014 no restarts needed, no stuck calls",
+          "Background utility calls enforce a single strict time budget (probe + retry + transport)",
+          "Token usage from background utilities is now reported accurately (no more zeros in accounting)"
         ]
       },
-      productVersion: "1.0.42",
-      announcement: "sema 1.0.42 \u2014 background utility calls (/btw, session titles, recaps) now run through the engine's new side-query channel: zero session files left behind, tool access structurally impossible, and accurate token accounting."
+      productVersion: "1.0.43",
+      announcement: "sema 1.0.43 \u2014 hardening for the new side-query channel: engine rollbacks/upgrades are now detected live (no stuck utilities), strict time budgets, and real token accounting flows through every background utility call."
     };
   }
 });
@@ -530605,16 +530592,20 @@ async function semaEngineNonStreamingLeg({
   if (options.outputFormat?.schema) {
     objective += side.jsonDisciplineSuffix(options.outputFormat.schema);
   }
-  const text2 = await side.engineUtilityText({
+  const r = await side.engineUtilityTask({
     objective,
     signal,
     maxTokens: options.maxOutputTokensOverride,
+    cheapSlot: true,
     // 池内模型透传(A2 兜底):queryWithModel/advisor 显式选池模型时按所选路由;CC 名走 cheap/默认
     modelRef: await side.resolvePoolModelRef(options.model)
   });
-  if (text2 === null) return null;
+  const text2 = r.text;
   const content = options.outputFormat?.schema ? side.extractJsonObjectText(text2) ?? text2 : text2;
-  return createAssistantMessage({ content });
+  return createAssistantMessage({
+    content,
+    ...r.usage ? { usage: r.usage } : {}
+  });
 }
 async function queryModelWithoutStreaming({
   messages,
@@ -609355,8 +609346,8 @@ function _temp161() {
 var import_compiler_runtime316, import_react314, SETTINGS_ERRORS_NOTIFICATION_KEY;
 var init_useSettingsErrors = __esm({
   "build-src/src/hooks/notifs/useSettingsErrors.tsx"() {
-    import_compiler_runtime316 = __toESM(require_compiler_runtime(), 1);
-    import_react314 = __toESM(require_react(), 1);
+    import_compiler_runtime316 = __toESM(require_compiler_runtime());
+    import_react314 = __toESM(require_react());
     init_notifications();
     init_state();
     init_allErrors();
@@ -655599,7 +655590,7 @@ Auth: unix socket -R \u2192 local proxy`, "info");
         pendingHookMessages
       }, renderAndRun);
     }
-  }).version("sema 1.0.42", "-v, --version", "Output the version number");
+  }).version("sema 1.0.43", "-v, --version", "Output the version number");
   program2.option("-w, --worktree [name]", "Create a new git worktree for this session (optionally specify a name)");
   program2.option("--tmux", "Create a tmux session for the worktree (requires --worktree). Uses iTerm2 native panes when available; use --tmux=classic for traditional tmux.");
   if (canUserConfigureAdvisor()) {